17th International Scientific and Technical Conference “FROM IMAGERY TO DIGITAL REALITY: ERS & Photogrammetry”

Phase One iXU-RS1000 Accuracy Assessment Report Yu. Raizman, PhaseOne.Industrial, Israel

1. Introduction The main steps of the project were executed by The Phase One of the several geospatial, geodetic and photogrammetric series iXU-RS1000 are well known worldwide companies, which are most experienced in the and very popular for small and medium size area relevant fields: mapping projects, corridor mapping, LiDAR 1. Test field preparation mapping, urban mapping, 3D City modeling and a. Planning – Dr. Yuri Raizman, PhaseOne. oblique imagery capturing, constructions and Industrial (http://industrial.phaseone.com/) infrastructure monitoring and inspection. b. Geodetic measurements – ARMIG Geodetic The 100MP cameras with size of 4.6 µ, Engineering Ltd. (www.armig.co.il/english) very high image capture rate -1 frame every 0.6 2. Test flight planning, flight execution and seconds and exposure time of up to 1/2500, a set of image preparation– Oodi Menaker, PhaseOne. metric lenses with different focal lengths (50, 70, Industrial 90, 110, 150 mm), provide a very effective solution 3. Image matching (automatic tie points in many areas of aerial mapping, monitoring and measurements), bundle block adjustment, GCP & object inspection. ChP measurements – Dr. Ziv Shragai from Simplex A very small form size (10x10x20cm including Mapping Solutions Ltd. (www.simplex-mapping. lenses) and a very light weight (less than 2 kg) are com) and Dr. Erwin Kruck from GIP, Dr. Kruck & also significant advantages of the – it may Co. GbR (http://bingo-atm.de/) be installed easily in every small and light aircrafts, 4. Stereoscopic GCP&ChP measurements – gyrocopters, medium size drones or UAVs that Mapping Technology Ltd. essentially increase a range of airborne vehicles 5. Accuracy analysis and report preparation – utilized for mapping and significantly reduces Dr. Yuri Raizman, PhaseOne.Industrial operational costs of mapping projects. The main goal of the report is photogrammetric The following software was used for the project: accuracy analysis and assessment of IXU-RS1000 • iX Capture – original image processing and camera. export to TIFF format • Agisoft Photoscan Pro – automatic image 2. Testing procedure matching and GCP & ChP measurements The photogrammetric accuracy assessment was • Bingo – aerial triangulation, bundle block done according to widely accepted aerial survey adjustment, camera calibration, and accuracy camera testing procedure, including the following assessment steps: • AtlasKLT – stereoscopic measurements of • Test field establishment GCP & ChP • Flight planning and flight execution • Automatic image matching and aerial 3. Test Field triangulation (AT) 3.1. Area • Flight camera calibration based on AT The test field is situated in the area of a village • AT accuracy analysis with different GCP & Kfar-Vitkin, Israel. It extends on 2.0 km in the ChP configuration West-East direction and 1.2 km in the South-North • Stereoscopic measurements of GCP & ChP direction. It is mostly an urban area with low height • Final conclusions one-two-story buildings.

33 October 16-19, 2017, Hadera, Israel

Figure 1: Test field The area characterized by existence of many as signalized GCPs. All GCPs are located on the permanent, visible from all sides and well-defined ground. The following manmade features were manmade features, which were chosen to serve mainly used as GCPs:

There are 53 GCPs in the test field. by two independent half-hour long observation sessions. 3.2. Geodetic measurements Before making the field measurements, all The following accuracies of the GCPs were GCPs were identified and marked on the Phase One received after processing all the observations: images of the area. The geodetic observations were RMSxy (cm) RMSz (cm) made according to static GPS survey procedure CSAR ± 0.3 ± 0.5 with one reference station 903AGR. The reference 903AGR vs CSAR ± 0.1 ± 0.1 station was established at the high and open area GCPs internal between ± 0.7 ± 1.2 in the north part of the test field. The reference observations station was measured against CSAR permanent GCPs absolute ± 0.8 ± 1.3 GPS station by two independent 1-hour long observation sessions. Every GCP was measured

34 17th International Scientific and Technical Conference “FROM IMAGERY TO DIGITAL REALITY: ERS & Photogrammetry”

4. Flight Planning 4.1. Aerial camera

The aerial camera iXU-RS1000 with of 90 mm was used for the test. The main parameters of the camera are as following:

Camera Specifications Lens type Rodenstock Focal length (mm) 90 FOV (across flight, deg) 33.0 FOV (along flight, deg) 25.1 f/5.6 Exposure principle 1 Exposure (sec) 1/2500 to 1/125 Image capture rate (sec) 0.6 Light Sensitivity (ISO) 50-6400 Dynamic Range (db) >84 Sensor Specifications Sensor type CMOS Pixel size (µm) 4.6 Array (pix) 11,608 x 8,708 Array volume (MP) 100 Analog-to-digital-conversion (bit) 14 Frame / Image Specifications Frame geometry Central projection Typical image size (MB) for TIFF 300 Image format PhaseOne RAW, Undistorted TIFF, JPEG Frame Coverage Frame width for 10 cm GSD (m) 1,161 Frame height for 10 cm GSD (m) 871 Frame area for 10 cm GSD (sq.km) 1.01

4.2. Aircraft installation Aircraft Cessna 172 was used for the flight. The camera was installed in the rear part of the plane with the following parameters of lever arm (distance between GPS antenna and the exit pupil of the camera):

dX (cm) -15.4 dY (cm) 10.1 dZ (cm) 0.99

35 October 16-19, 2017, Hadera, Israel

4.3. Aerial survey parameters • Side overlap – 49% • Forward overlap – 80% The flight was planned and executed with the • Frame size – 450m x 340m following aerial survey parameters: • Orthophoto angle - 17° • Flight altitude (above ground) - 2,500 feet • Building lean – 15% • GSD – 4 cm • Ground speed – 100 knot • Distance between flight lines – 230 m • Strips: SN – 9; WE – 2;

Figure 2: Test field with flight lines 5. Image matching and aerial triangulation residuals on GCPs, which are not participated in the adjustment – Check Points (ChP). The image matching was performed with Agisoft Photoscan Pro software. There were There were five runs of Bingo for different 202 images in the matching and 6,434 tie points purposes: (39,335 x,y image measurements) were identified 1. Tie point and all 53 GCP adjustment with during the matching. The image coordinates of all additional parameters 61, 62 for camera calibration GCPs were manually measured in Photoscan Pro. 2. Tie point and all 53 GCP adjustment with After the matching, all tie points and GCPs additional parameters 17,18,19,39,61,62 and 65 image coordinates were provided to Bingo software for camera calibration for further processing. 3. Tie points, 5 GCP and 48 ChP for the block geodetic accuracy estimation The Bingo software was used for bundle-block 4. Tie points, 9 GCP and 44 ChP for the block adjustment with additional parameters for self- geodetic accuracy estimation calibration and exterior orientation of the block 5. Tie points, 15 GCP and 38 ChP for the block with different number and configuration of used geodetic accuracy estimation GCPs. Photogrammetric accuracy of the block is estimated by residuals on tie points or by sigma Common parameters of the block for the above 0. Geodetic accuracy of the block is estimated by Bingo runs are as following:

36 17th International Scientific and Technical Conference “FROM IMAGERY TO DIGITAL REALITY: ERS & Photogrammetry”

• Number of images – 202 • Number of used tie points – 3600 The following graph presents a tie points • Number of skipped tie points - 2835 distribution per images: • Mean number of points per image – 222 • Number of measured GCPs - 53

Images in the middle of the block with multiple images along the perimeter of the block with only overlaps have more identified tie points than a few overlapping images.

6. Camera self-calibration analysis participated in the adjustment. Only approximate GPS data was used for the adjustment. The self-calibration procedure is a simultaneous bundle adjustment of all images of the block with The camera self-calibration procedure was many tie points and many GCPs participating in processed twice with different sets of additional the adjustment. During the self-calibration, the parameters. The first set included additional camera interior orientation parameters (camera parameters 61 and 62, which correspond to K1 model), F, xo, yo, k1, k2, p1, p2, p3 are calculated. and K2 radial distortion coefficients in standard The accuracy of these parameters depends on Brown-Conrady distortion model. many factors – number and size of image overlaps, The second set included additional parameters number and accuracy of identified tie points, 17, 18, 19, 39, 61, 62 and 65 and was provided and number and accuracy of GCPs, existence and tested by Dr. Erwin Kruck, the developer of Bingo accuracy of GPS during the flight and some more. software.

As mentioned above, 202 images with an After the adjustment of all images, the following average 202 tie points per image and 53 GCPs results obtained:

37 October 16-19, 2017, Hadera, Israel

Tie points residuals 57 GCP Image Ground Ground dx (µm) dy (µm) dX (cm) dY (cm) dZ (cm) dX (cm) dY (cm) dZ (cm) MIN -7.1 -8.1 0.3 0.3 0.5 -3.0 -3.1 -0.3 MAX 7.2 9.3 5.7 3.5 16.6 1.2 1.7 0.3 Mean 0.0 0.0 0.7 0.6 3.2 0.0 0.0 0.0 STDEV 1.7 1.6 0.4 0.3 2.1 0.6 0.7 0.1 RMS 1.7 1.6 0.8 0.7 3.8 0.6 0.7 0.1 Sigma 0 1.79 / 1.83 µm (depending on a number of additional parameters) 0.4 of pixel size

The results clearly present a very high F (mm) Xp (mm) Yp (mm) photogrammetric accuracy of the block. The RMS 89.5447 0.0244 0.0149 of tie point s is just 1.7 µm in the image plane (0.3 RMS (µm) 10.1 1.5 1.5 of pixel). The RMS of tie points on the ground is RMS (pix) 2.20 0.33 0.33 0.8 cm in position (0.2 of pixel size on the ground) and 3.8 cm in height that is 0.8 of pixel size on the The low accuracy of the focal length calculation ground. (10.1 µm) may be explained by the low accuracy The relatively large difference in position and of GPS data already mentioned above. height accuracy may be explained by two factors: low accuracy of GPS during the flight and long During the self-calibration the camera model focal length (90 mm) of the camera that influence distortion was also calculated. The left picture the height accuracy. below presents a clearly visible symmetrical radial distortion model of the camera with a maximal After the adjustment, the following main camera distortion of 24 µm before fixing the image. parameters were calculated:

38 17th International Scientific and Technical Conference “FROM IMAGERY TO DIGITAL REALITY: ERS & Photogrammetry”

The right image presents the same image after its fixing with 61, 62 additional parameters describing a simple symmetrical radial distortion model (corresponds to K1 and K2 coefficients). Maximal residual here is 0.91 µm, which is just 0.2 of pixel (4.6 µm). The third picture from the left presents the same camera after implementing the following additional parameters 17,18,19,39,61,62,65, describing more complicated distortion model. In this case, the maximal residual equals to 0.86 µm. As a conclusion, should be explicitly stated that the distortion model of the camera iXU-RS1000 with 90 mm focal length fully corresponds to a standard Brown-Conrady distortion model and images captured with the camera may be easily transformed to undistorted model with a maximal residual less than 1 µm.

7. Accuracy analysis for different GCP & ChP configuration were chosen for testing geodetic accuracy of the block. Several different GCP and ChP configuration One of them is based on 15 GCP and 38 ChP:

39 October 16-19, 2017, Hadera, Israel

The following results were obtained for this configuration: 15GCP 38 Check Points dX (cm) dY (cm) dZ (cm) dX (cm) dY (cm) dXY (cm) dZ (cm) MIN -0.8 -1.2 -0.2 -6.3 -6.2 8.8 -8.1 MAX 0.6 1.8 0.2 3.3 5.9 6.8 5.3 Mean 0.0 0.0 0.0 -0.3 0.1 0.3 -1.8 STDEV 0.4 0.6 0.1 1.6 2.1 2.7 4.0 RMS 0.4 0.6 0.1 1.7 2.1 2.7 4.4

With 15 GCP and for 38 Check Points the The following chart is a graphical representation RMSxy = ± 2.7 cm (0.7 pix) and RMSz = ± 4.4 cm of the residuals on Check Points: (1.1 pix) were obtained.

The red vectors represent XY residuals and GCPs used for adjustment and exterior orientation the blue one – Z residuals. There is no visible of the block. systematic errors in the block. The altimetric accuracy of the block on Check Points varies from 1.0 to 1.5 pix depending on The planimetric accuracy of the block on the number of GCPs. This altimetric accuracy Check Points is always at the level of 0.7 pixel is considered as high accuracy and even may be independently on the number and configuration of improved by use of high accuracy GPS data.

8. Accuracy analysis from stereoscopic measurements

The aim of the stereoscopic test is to check a measurements may be calculated as flowing: potential accuracy of stereoscopic measurements on images captured by iXU-RS1000 camera Mxy= H/F*mxy ; Mz=H/b*mp with focal length of 90 mm and to test a possible existence of the vertical parallax residuals after Where: relative orientation . Mxy – theoretical planimetric accuracy of stereoscopic measurement on the ground, The theoretical accuracy of stereoscopic Mz – theoretical altimetric accuracy of

40 17th International Scientific and Technical Conference “FROM IMAGERY TO DIGITAL REALITY: ERS & Photogrammetry”

stereoscopic measurement on the ground, Several measurements were done with stereopairs mxy – accuracy of x,y measurement on the formed with images from adjacent strips with image plane, using side overlap for stereoscopy. Every point mp – accuracy of parallax measurement on the was measured twice on the same stereopair. The image plane. Atlas KLT software was used for the stereoscopic measurements. With the assumption, that mxy=mp=0.3 pix, for camera with F=90 mm and forward overlap of Exterior orientation parameters were calculated 60% the theoretical planimetric accuracy will be from bundle block adjustment with self-calibration 0.3 pix and the altimetric – 1.7 pix. and with use of all 56 GCPs for the block orientation. It was done to eliminate an impact Mapping Technology Ltd. executed stereoscopic of possible block deformations on stereoscopic measurements. For the analysis, 56 GCPs on measurements. 91 stereopairs were measured. Stereopairs with 60% forward overlap were mainly selected. The results of the test are as following: Xg-Xp (cm) Yg-Yp (cm) Zg-Zp (cm) Xg-Xp (pix) Yg-Yp (pix) Zg-Zp (pix) MIN -5.10 -6.40 -13.20 -1.27 -1.60 -3.30 MAX 5.90 6.30 18.20 1.48 1.58 4.55 Mean -0.16 -0.18 2.94 -0.04 -0.04 0.73 STDEV 2.03 2.40 6.87 0.51 0.60 1.72 RMSE 2.04 2.41 7.47 0.51 0.60 1.87 The distribution of dZ residuals may be presented as following:

9. Conclusions camera’s capabilities to be used in high accuracy mapping projects. Only the camera with focal This Phase One iXU-RS1000 Accuracy length of 90 mm was tested. Assessment Report was dedicated to check the

41 October 16-19, 2017, Hadera, Israel

The test preparation included: 1. Phase One iXU-RS1000 is a metric camera 1. Test field development with precisely defined with stable and clearly definable interior orientation and measured GCPs, parameters, 2. Tie points automatic measurements (image 2. The images captured by the camera are of matching) and GCP measurements with Agisoft high geometric and radiometric quality ensuring Photoscan Pro software, the use of the camera in high accuracy mapping 3. Bundle block adjustment with Bingo software. projects, including sterecompilation. 4. Stereoscopic measurements with Atlas KLT 3. The final photogrammetric accuracy without digital photogrammetric workstation. use of GNSS data reached 0.5 pixel in position and 1.0 pixel in altitude. The results obtained during the data analysis 4. The right use and high quality of GNSS data clearly evidence that: will even improve the results.

42